Tyler Cowen recently challenged us to stump an AI named “o1”.
1. o1 is a very good economist. Try cutting it into a stump if possible.
I couldn’t find o1, so I used ChatGPT 4o mini. I hope someone will try the following question on o1 and report the results in the comments section.
What I’ve always noticed about AI is that it’s like a very smart student who can memorize vast amounts of information but lacks the ability to really see the essence of the problem. is. In the example above, I gave hint after hint, essentially begging the AI to avoid inferring from price changes. Little did they realize what they were doing, they ended up telling me that the market was in equilibrium even though Qs was rising and Qd was falling. It’s gone.
Maybe I used the wrong AI, or maybe this question was unusually complex. (For college students, “difficult” means problems that require you to actually understand something and can’t just repeat what you wrote in your notes.)
So let’s take the question Mr. Tyler asked the o1 economists.
How current macroeconomists find it so difficult to predict the dynamics of inflation, and why they use different forms of monetarism, even if they do not understand what is going on behind the movement of the price level. Write an essay about how you came to reject it as even an approximation of it.
The answer is quite long so I will focus on the important parts of the answer.
Several factors contribute to the increasing complexity and unpredictability of inflation. First, financial innovations and regulatory changes in the late 20th and early 21st centuries dramatically changed the relationship between money and economic activity. Monetary aggregates such as M1 and M2, which once served as reliable indicators of policy stance and future inflation, are now increasingly affected by changes in the speed at which money moves, the prevalence of shadow banking, and the globalization of financial flows. , the behavior is unstable. Simply put, where money resides and how quickly it moves through the economy has become too fluid and complex to be captured by the simple rules of the old monetarists.
Second, the nature of central banks and fiscal policy making has changed. Central banks are currently intervening in a variety of non-traditional ways, from large purchases of financial assets to forward guidance on policy expectations. These tools cannot be adequately explained in terms of the classical monetarist perspective, which focuses on controlling specific measures of the money supply. The recent experience after the global financial crisis illustrates this clearly. The Federal Reserve and other central banks implemented unprecedented quantitative easing programs and dramatically expanded their balance sheets. According to traditional monetarist logic, this rapid increase in the monetary base should have caused significant inflation. But inflation has remained below target levels in many developed countries for years, confounding those who relied on the old monetary aggregate index.
Third, determining prices today involves a bewildering interplay of global supply chains, technological advances, labor market transformations, and changes in consumer behavior. Globalization means that the prices of goods and services are affected not only by domestic financial conditions, but also by distant supply shocks, currency fluctuations, and international competition. Technological innovations can increase productivity and compress prices in certain sectors, while reducing the impact on other parts of the economy. Labor markets are also evolving, with changes in trade union power, demographic changes and changing labor participation patterns affecting wage formation and cost pressures. These micro-level frictions and structural changes make the old macro-level equations linking money supply growth and inflation too crude and imprecise.
Sorry, this is mostly a combination of empty platitudes and irrelevant comments. Again, this is the kind of thing where I expect students to stop if they don’t know the answer. I’ll explain why later, but first let me clarify an important point. I’m pretty sure o1 is smarter than me, at least in terms of how most people define “intelligence.” In other words, o1 knows a lot more than I do. The New York Times is much more intelligent than I am, in the same sense that The New York Times is much more intelligent than I am. It’s really, really impressive.
But o1 is not a good economist. Sorry, that’s not the case.
First, let’s get rid of the last paragraph. This is complete nonsense. Technology has been advancing for over 100 years, but the pace of productivity growth has not accelerated. Therefore, it is not a factor that makes inflation difficult to predict. The strength of unions waxes and wanes, but that doesn’t make inflation any harder to predict. Changes in consumer behavior? Students are really desperate now. This is a “I need to fill the page with something” kind of comment. Global supply chains?Again, inflation is not difficult to predict. Not at all. World commodity prices had a greater impact on U.S. inflation 100 years ago than they do today. None of these four factors makes forecasting inflation any more difficult. In fact, I don’t know that inflation is becoming increasingly difficult to predict. When I was younger, I think it was much more difficult to predict inflation than it is now.
The first sentence of the middle paragraph isn’t too bad. Central banks have really changed in some respects. And the relationship between base and aggregate is really loose. Unfortunately, the AI doesn’t seem to know why the relationship has become so tenuous. It talks a lot about irrelevant things like quantitative easing and misses the important point that interest payments on reserves and the zero lower bound problem make the money multiplier much more unstable. AI also says fiscal policy decisions have changed. Perhaps that was just a wild guess. It is hard to imagine any changes to fiscal policy that would make inflation difficult to predict. Given the AI’s silence on this issue, I suspect it also doesn’t understand what has changed in fiscal policy or why that change makes inflation harder to predict.
In the first paragraph the AI says:
Monetary aggregates such as M1 and M2, which once served as reliable indicators of policy stance and future inflation, are now increasingly affected by changes in the speed at which money moves, the prevalence of shadow banking, and the globalization of financial flows. , the behavior is unstable.
error. The aggregate value has never been a reliable indicator of inflation. The velocity of money is always unstable. However, don’t textbooks say that the velocity of money was stable in the past? Yes, many people do. Perhaps that’s where the AI got the wrong information. It may be more unstable now, but the aggregation has never been stable enough to be a good predictor.
Also keep an eye out for strange lists including velocity, shadow banking, and fund flows. That’s kind of ridiculous. An AI that knew what it was talking about would put a period after the speed and write another sentence explaining why the speed changed. At this point, the concepts are jumbled together and confusing.
If you are a math-oriented person who is confused by economics, the following formula may help you understand my point.
V = year/month
The AI is basically saying that the PY/M ratio is changing due to changes in V, shadow banking, capital flows, etc. The ratio is V!!Other variables can help explain why V changes.
When I stumble upon an article in the NYT in an area in which I have some expertise, the paper suddenly seems less intelligent, even though the NYT is much smarter than I am. I don’t call them “stupid,” because even in the worst-case scenario, the NYT employs talented reporters. And we cannot expect them to devote their lives to the study of financial economics.
In my view, the o1 answer provided by Mr. Tyler is slightly below the level of the NYT. At least the Times would have mentioned reserve interest and the zero lower bound. Nevertheless, it is truly amazing when it comes to the amount of information that AI remembers. Even in the field of financial economics, AI often has more information at its fingertips than I do. AI could write a better essay than I could about all the Fed’s new policy tools. It’s kind of an understanding of where that fits. I don’t even know if I can express this concept in words. Perhaps understanding doesn’t really exist, and we just think we understand something deeply. Perhaps a superintelligence would scoff at my writing the same way I scoff at an AI that makes inferences from price movements.
All I know is that AI has a long way to go before it impresses me in the area of understanding financial economics. But outside of my field of expertise, I’m already very impressed.
The AI Overview defines Germanic amnesia as follows:
Germanic amnesia (GMA) is when a person reads a news article about a familiar subject and perceives it to be error-prone, yet continues to read the rest of the newspaper as if it were more accurate. It’s a defense mechanism that sometimes occurs. . This term was coined by Michael Crichton after discussion with Murray Gell-Mann.
